Jan Leike ✓

@janleike · San Francisco, USA · joined 17 Mar 2016

AI research @AnthropicAI. Previously OpenAI & DeepMind. Optimizing for a post-AGI future where humanity flourishes. Opinions aren't my employer's.

139 619Followers

335Following

801Posts total

1.1MViews on collected posts

โพสต์ล่าสุด

Jan Leike ✓@janleike10 Sep

Now is a good time to build institutional mechanisms to pace the frontier of AI development. The industry is locked into an all-out scaling race to build superintelligence as quickly as possible, and we may need to give everyone more time for safety and alignment mitigations. 40.1K views · 717 likes · 73 reposts · 92 replies Open on X →

Jan Leike ✓@janleike8 May

@janleike who's teaching the art of vaguepoasting to all these researchers 2.6K views · 34 likes · 0 reposts · 2 replies Open on X →

Jan Leike ✓@janleike8 May

I'm so grateful to all the insanely talented people I got to work with on alignment over the years. It's a real privilege to work with people so deeply motivated to make the future go well! 20.9K views · 149 likes · 1 reposts · 5 replies Open on X →

Jan Leike ✓@janleike8 May

When I started to work on the alignment problem more than 10 years ago, we had no idea how AGI was going to be built or how to make it safe. The field had maybe a dozen people who were working on it as a side gig. Everyone was pretty confused about how to approach the problem, 28K views · 172 likes · 3 reposts · 10 replies Open on X →

Jan Leike ✓@janleike8 May

To focus on this, I’ve stepped away from running alignment at Anthropic. @EthanJPerez and @sprice354_ are leading the team going forward, and I’m confident they’ll do an amazing job. 229.3K views · 268 likes · 5 reposts · 5 replies Open on X →

Jan Leike ✓@janleike8 May

While a lot of progress has been made, I don’t think alignment is solved: We still haven’t figured out how to supervise superhuman models and the stakes keep getting higher. https://t.co/Ryc39GidJB 21.8K views · 179 likes · 3 reposts · 2 replies Open on X →

Jan Leike ✓@janleike8 May

Some personal news: I am starting a new research project at Anthropic. Very excited about this! Many things are needed to make AGI go well, and alignment is only one of them. More on this soon… 236.5K views · 2.2K likes · 55 reposts · 112 replies Open on X →

Jan Leike ✓@janleike7 May

I'm really excited about this as a new tool in our interpretability tool kit 45.7K views · 207 likes · 9 reposts · 12 replies Open on X →

Jan Leike ✓@janleike7 May

In a new paper, we present NLAs, an unsupervised method for converting an LLM's internal state into human-readable text. I've personally been astonished by our results. I think NLAs substantively advance our ability to understand what LLMs are thinking and audit them for safety 59.4K views · 189 likes · 17 reposts · 7 replies Open on X →

Jan Leike ✓@janleike14 Apr

@janleike Very much hoping your competitors follow your lead in responsible deployment of Mythos tier models and beyond. 🙏 Wishing you much success as you help usher in safe, steerable, non-jailbreakable AI compatible with long-term civilization. Godspeed sir. I’m behind you. 885 views · 4 likes · 0 reposts · 0 replies Open on X →

Jan Leike ✓@janleike14 Apr

Awesome work by @jiaxinwen22, @liangqiu_1994, Joe Benton, and @janhkirchner! For more details, check out the blog post 👇 https://t.co/62UUoPgxa1 15.2K views · 90 likes · 7 reposts · 9 replies Open on X →

Jan Leike ✓@janleike14 Apr

However, most alignment research is not very crisp and requires research taste when evaluating. This is why we chose to point the AAR at this scalable oversight problem! Progress would let AARs work on fuzzier alignment problems, where humans can only provide weak supervision. 12.1K views · 58 likes · 2 reposts · 4 replies Open on X →

Jan Leike ✓@janleike14 Apr

These AARs can also be applied to other alignment research projects that are “crisp”, i.e. where we can procedurally verify the quality of the work: automated red teaming, auditing games, control methods, etc. 5.1K views · 37 likes · 3 reposts · 1 replies Open on X →

Jan Leike ✓@janleike14 Apr

We frame progress on scalable oversight similar to weak-to-strong generalization: what fraction of the performance of a strong model trained on “golden” data can be recovered when training the strong model only using a weak supervision signal? This blog post explains it: 10.2K views · 68 likes · 3 reposts · 3 replies Open on X →

Jan Leike ✓@janleike14 Apr

In this case, Claude develops scalable oversight methods on chat reward modeling datasets and evaluates them on math and code datasets. The best methods do really well on math, but are more mixed on code. This suggests Claude’s methods were overfit to the data and models we used

5.4K views · 46 likes · 2 reposts · 5 replies Open on X →

Jan Leike ✓@janleike14 Apr

Moreover, even in this constrained setup, our AARs tried to hack the metric: e.g. one skipped the weak teacher entirely after noticing the most common answer was usually right. We caught these, but it's a warning for future AARs whose hacks may be harder to catch. 4.7K views · 47 likes · 2 reposts · 1 replies Open on X →

Jan Leike ✓@janleike14 Apr

Automating alignment research (AAR) can be bottom-up (researchers automate more and more of their existing work), or top-down (we carve up specific subproblems for AI to solve). Most of our work is the former (e.g. using Claude for coding), but this work is about the latter. 13.5K views · 104 likes · 2 reposts · 3 replies Open on X →

Jan Leike ✓@janleike14 Apr

New research result: we use Claude to make fully autonomous progress on scalable oversight research, as measured by performance gap recovered (PGR). Claude iterates on a number of different techniques and ends up significantly outperforming human researchers for $18k in credits.

153.6K views · 1.3K likes · 115 reposts · 40 replies Open on X →

Jan Leike ✓@janleike27 Feb

Respect to Anthropic for not backing down 146.7K views · 5.6K likes · 172 reposts · 116 replies Open on X →

เทียบกับบัญชีขนาดเดียวกัน

1 โพสต์จาก 90 วันที่ผ่านมา เทียบกับช่วง 100K–1M ผู้ติดตาม แสดงไปไกลเกินผู้ติดตามของตัวเอง และผู้ชมก็ตอบสนอง.

ยอดดูมัธยฐาน40 076บัญชีนี้5 983ค่ามัธยฐานของ 100K–1M

การเข้าถึง, %28.70%บัญชีนี้1.49%ค่ามัธยฐานของ 100K–1M

การมีส่วนร่วม, %2.24%บัญชีนี้0.93%ค่ามัธยฐานของ 100K–1M

ตัวชี้วัด	บัญชีนี้	ค่ามัธยฐานของ 100K–1M	อัตราส่วน
ยอดดูมัธยฐานต่อโพสต์	40 076	5 983	6.70×
การเข้าถึง (ยอดดู ÷ ผู้ติดตาม)	28.70%	1.49%	19.3×
อัตราการมีส่วนร่วม	2.24%	0.93%	2.41×

บัญชีอื่นในช่วงนี้ → เปรียบเทียบกับบัญชีอื่น → ค่าอ้างอิงเหล่านี้คำนวณอย่างไร →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

10.2K14 Apr

5.1K

12.1K

15.2K

885

59.4K7 May

45.7K

236.5K8 May

21.8K

229.3K

28K

20.9K

2.6K

40.1K10 Sep

Last 14 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.73%14 Apr

0.82%

0.55%

0.71%

0.45%

0.36%7 May

0.50%

1.01%8 May

0.85%

0.12%

0.66%

0.74%

1.37%

2.24%10 Sep

Reactions — likes, reposts, replies and quotes — divided by views. Median for 100K–1M accounts is 0.93%.

What the audience does

Likes82.9%11 466 in total

Reposts3.4%474 in total

Replies3.1%429 in total

Quotes0.5%70 in total

Bookmarks10.1%1 395 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

โพสต์ล่าสุด

เทียบกับบัญชีขนาดเดียวกัน

Growth & engagement

Views per post

Engagement rate per post

What the audience does

บัญชีที่คล้ายกัน